meng-fan/ontram-paper

Repository accompanying arXiv:2010.08376

Ordinal Neural Network Transformation Models (ONTRAM)

This repository accompanies arXiv:2010.08376

Contents:

Experiments: Contains experiments to reproduce all results from the paper in R and Python.

• wine

  • data: Contains the wine quality (red) data split into the 20CV folds mentioned in the paper. The data was taken from here.
  • learning-efficiency: Code for reproducing Figure 8
  • models: Code for fitting all models for the wine data listed in Table 1 (MCC, QWK, POLR, CI_x, SI-LS_x, SI-CS_x^*)
  • permuted-class-labels: Code for reproducing Figure A1

• UTKFace

  • models: Python notebooks for fitting the models listed in Table 1 (MCC, MCC-x, QWK, SI-LS_x, CI_B, CI_B-LS_x, SI-CS_B, SI-CS_B-LS_x) and the models to reproduce Figure 11.
  • simulate-tabular: Code for simulating the tabular predictors as illustrated in Figure 5.

Miscellaneous: Miscellaneous scripts illustrating scoring rules and results from the paper.

• qwk-impropriety.R: R-script that computes the numerical example from Appendix D to show impropriety of the QWK loss.

The ontram R package:

• The R package implementing ordinal neural network transformation models lies in a separate repository and can be installed from within R via remotes::install_github("LucasKookUZH/ontram-pkg").